5

What is the newest MIP or CP solver (which is model based exact solver)?

I want to follow and wonder developments on solvers. But finding new developed solvers (experimental, academic, commercial etc.) with search is not enough, and sometimes can be missed. Answers to this question can be guiding.

kur ag
  • 815
  • 5
  • 14

2 Answers2

4

Below are some popular solvers for MIP & CP (Not an exhaustive list though)

  1. MIP Solvers: Gurobi, CPLEX, XPRESS(FICO), CBC, SCIP, etc..
  2. CP Solvers: MiniZinc, OR-Tools, CP Optimizer(CPLEX), etc..

One of the newest solver that I can across recently for LP/MIP/QP was HIGHS.

anjikum
  • 984
  • 4
  • 8
  • 2
    Note that MiniZinc is not a CP solver -- it is a high level language which can generate input for other other CP (and MIP and SAT) solvers. – Chris Jefferson Feb 01 '22 at 09:53
3

OptaPy might be the newest solver at the time of writing.

OptaPy is an open source Python solver that uses parts of OptaPlanner inside (but it's considered a different solver). It was announced on 5 October 2021.

Geoffrey De Smet
  • 4,851
  • 10
  • 34
  • But it is not a CP solver. It is a heuristic. – Erwin Kalvelagen Jan 31 '22 at 08:00
  • Good point, fixed. – Geoffrey De Smet Jan 31 '22 at 08:04
  • 2
    It may not even be a solver: it does not know when a problem is solved. – Erwin Kalvelagen Jan 31 '22 at 08:08
  • I guess none of the answers on this thread so far are wrong :) None of them are (newest and (CP or MIP)). – Geoffrey De Smet Jan 31 '22 at 08:09
  • @ErwinKalvelagen Then by that logic all metaheuristics solvers aren't solvers? Do note that OptaPy and OptaPlanner support exhaustive search too, so then they do know when an optimal solution is reached. – Geoffrey De Smet Jan 31 '22 at 08:10
  • 5
    I think the name "solver" implies that it knows when an optimization problem a solved. So no, a heuristic in my view is not a solver. It does not really "solve" a problem. Unless you restate the problem as "give me a good solution for ..." (where good really means "not bad"). There is just no concept of optimality in (most) heuristics. – Erwin Kalvelagen Jan 31 '22 at 08:13
  • Intresting point of view. Would you consider OR-Tools's VRP solver or LocalSolver's solver as a solver? – Geoffrey De Smet Jan 31 '22 at 08:14
  • 4
    OR-Tools and LocalSolver can find guaranteed optimal solutions. Yes, they are proper solvers. They solve optimization problems. They know when they are done and the problem is solved. – Erwin Kalvelagen Jan 31 '22 at 09:56
  • 1
    I agree with @GeoffreyDeSmet here, I don't consider the notion of solver to be related to optimality, but rather to finding solutions or bounds. Would IPOPT stop being a solver when executed on a non-convex problem? In addition, MILP and CP solvers are often used just as heuristics – fontanf Jan 31 '22 at 10:06
  • 1
    IPOPT solves non-convex problems to local optimality. It knows when the problem is local optimal and it can stop then and tell the user it is solved. – Erwin Kalvelagen Jan 31 '22 at 10:10
  • 3
    You should add a disclaimer acknowledging your bias when advertising your own projects. – Robert Bassett Jan 31 '22 at 13:44
  • 1
    Robert, see my profile picture or profile description. If LocalSolver, Laurent Perron (from OR-tools) and me (from OptaPlanner) have to add a disclaimer to every post we do here, it's cumbersome. In none of our cases, our affiliation is a secret. – Geoffrey De Smet Jan 31 '22 at 18:35